muelea / buddi

Our model BUDDI learns the joint distribution of interacting people
https://muelea.github.io/buddi/
MIT License
140 stars 12 forks source link

Evaluation on Hi4D? #10

Open noahcao opened 7 months ago

noahcao commented 7 months ago

Thanks for the wonderful work!

I saw that you reported the evaluation results on Hi4D. However, I could not find a way to reproduce that given your instructions in README? Could you help to indicate how to reproduce the reported Hi4D performance as in the paper? Thanks!

muelea commented 7 months ago

Hi noahcao,

you can run the Hi4D evaluation script with this command once the Hi4D SMPL-X fits become available. We're working on a solution to publish them in accordance with the Hi4D license.

You'd first need to run the optimization with BUDDI on Hi4D or download the fits here and unpack them into PATH/TO/Optimization_with_BUDDI_Hi4D_test. Then you can run the evaluation script like this: python llib/methods/hhcs_optimization/evaluation/hi4d_eval.py \ --exp-cfg llib/methods/hhcs_optimization/evaluation/hi4d_eval.yaml \ --predictions-folder PATH/TO/Optimization_with_BUDDI_Hi4D_test --eval-split test